Qwen/Qwen2-Audio-7B-Instruct · Does the Qwen2-Audio-7B-Instruct model support Function Call?

I passed in a prompt following the function call format used for Qwen2-7B-Audio-Instruct, but the result did not return the expected function invocation.

<|im_start|>system\nYou\'re a helpful assistant\n\n# Tools\n\n## You have the following tools:\n\n### get_current_weather\n\nget_current_weather: Get the current weather in a given location Input parameters: {"type": "object", "properties": {"location": {"type": "string", "description": "The city and state, e.g., San Francisco, CA"}, "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}}, "required": ["location"]} The input to this tool should be a JSON object.\n\n## You can invoke the following command zero, one, or more times in the response to call the tool:\n\n✿FUNCTION✿: tool name, must be one of [get_current_weather].\n✿ARGS✿: tool input\n✿RESULT✿: tool result\n✿RETURN✿: Reply based on the tool result, and render images using ![](url)\n<|im_end|>\n<|im_start|>user\nAudio 1: <|audio_bos|><|AUDIO|><|audio_eos|>\nHow is the weather in Shanghai today?\n<|im_end|>\n<|im_start|>assistant\n

The result I got:

"Today's weather in Shanghai is clear with cool temperatures, so it's recommended to bring a jacket when going out."

Is this due to the audio model not being fine-tuned for function calls?

This translation should convey the issue accurately for an English-speaking audience.